AI's Next Fight Is Who Pays for the Grid

The House passed H.R. 9340, the Ratepayer Protection Act, 417–3 on September 16 under suspension. Sponsors Gabe Evans (R-Colo.) and Kathy Castor (D-Fla.) framed it as keeping residential bills from funding AI campuses. Energy and Commerce called it a first step and sent it to the Senate. The wire is true: Washington is now in the large-load file.

The residual is why the fights do not end if hyperscalers “pay their way.” If a campus pays the full incremental cost of the steel built to serve it, why do utilities still demand minimum bills, exit fees, collateral, and contracted-capacity floors?

PURPA tells states to consider, not to copy Ohio

The reported text amends Public Utility Regulatory Policies Act section 111(d). A rate for a large-load customer — IT load of 100 MW or more at a site or campus, contracted on or after enactment — “shall be designed to recover” the full incremental cost of generation, transmission, or distribution upgrades needed to serve that load, including if the customer terminates or stops buying. Before those upgrades, the utility “shall require” financial assurances or contributions covering their cost. States get one year to start considering the standard and two years to decide. Prior comparable state action can skip the homework.

That is a federal consider-standard, not a national tariff. Commissions still classify what is “incremental,” how to collateralize it, and whether a comparable local rule already counts. Democrats who wanted more teeth were not inventing a complaint. A 417–3 vote can still leave Oregon, Ohio, and a laggard ISO on different clocks. Time-to-power stays a queue and a transformer, not a House roll call. Chip quarters can look fine while the campus still has no electrons.

Unused megawatts are a second bill

Upgrade recovery pays for the line that got built. Take-or-pay pays for the megawatts reserved and not used — the stranded-utilization risk if a hyperscaler delays, downsizes, or walks after the utility sized the system to a contract.

AEP Ohio’s data-center tariff, effective 23 July 2025, is the 85 percent / 12-year specimen. New large loads face minimum demand that can sit at 85 percent of contract capacity, a ramp of 50–90 percent over as many as four years, an initial term of ramp plus eight years, collateral on financial viability, and an exit fee if the project cancels. Oregon’s PGE Schedule 96, Order 26-154, effective 10 June 2026, is harsher on some dials: 100 percent of distribution upgrades, demand charges at 90 percent of contracted capacity even if unused, contracts from 10 years to 30 years at 220 MW, exit fees on remaining demand and unspent distribution plant, and a 1¢/kWh surcharge above 100 MW.

Those are not accounting aliases for the same incremental-upgrade invoice. A customer can fund a substation and still owe monthly for headroom it never draws. That is why disputes continue after a “pay your own upgrades” slogan. Ghost racks already asked whether demand shows up. The tariff asks who pays when it does not.

Local tax and permit fights sit on a third axis: even a fully collateralized interconnection can fail a county vote. That does not socialize the transformer. It can still kill the site.

Incremental cost recovery pays for steel that was built. Minimum bills pay for megawatts that were reserved.

What still mis-prices is treating H.R. 9340 as a closed national rate or as proof that hyperscale IRRs are unchanged. This is not a simple replay of cheap dark fiber. The Senate can sit on the bill. If it becomes law, watch the two-year PURPA dockets for whether “comparable” means Ohio’s 85 percent floor or a weaker contribution letter. The test for a given campus is the executed large-load agreement: contracted MW, take-or-pay percent, collateral, and the exit schedule — not the House score.